Book Reviews: Japanese Phrase Structure Grammar: A Unification-Based Approach

Professor Gunji's book continues the tradition of explicitness in linguistic description so well demonstrated by Gazdar, Klein, Pullum, and Sag (1985) (GKPS). In theoretical orientation, too, there is a heavy debt to the latter. The details, however, show a certain trend towards the lexicalist position exemplified in theories such as head-driven phrase-structure grammar (Pollard and Sag 1987), and categorial unification grammar (Uszkoreit 1986, Karttunen 1986, Zeevat et al. 1986). All these theories have been developed with a strong computational orientation. GKPS could be considered essential reading for anyone intending or attempting to develop a serious computational grammar for English, and Gunji's Japanese Phrase Structure Grammar (JPSG) stands in the same relation to Japanese. In fact, JPSG is used as the basis of the natural-language component of the Japanese Fifth Generation Computer project. The book, a considerably revised version of the author's Ohio State M.A. thesis, presents a thorough and extensive analysis of the central syntactic phenomena of the Japanese language--control in complementation, reflexive and zero anaphora, unbounded dependencies, and word-order variation. After a brief introduction, the apparatus to be used in the analysis is presented. Despite the book's title, the phrase-structure rule component is reduced to a single schematic rule, M ~ D H. The burden of constraining the weU-formedness of the local trees licensed by this rule falls on several principles that govern the distribution of features in the categories M(other), D(aughter), and H(ead). These include the head-feature principle (HFP) and foot-feature principle (FFP), familiar from GKPS; an adjunct-feature principle, which requires the value of the ADJUNCT feature of the category D to unify with the category H; and a SUBCAT feature principle (SFP), which ensures the satisfaction of lexical complementation requirements, as in HPSG. SUBCAT is a feature that takes sets of case-marked categories as its value, and the SFP states that in complementation, the value of the mother's SUBCAT is that of the head minus the element that unified with the non-head daughter. Gunji assumes for most of the book a semantics with /3-reduction of A-calculus expressions as the basic combination operation, and it is this that is explained in the preliminaries section. However, in the final chapter, on word order, he suggests that a semantic combination based on unification is more appropriate for handling a language with the word-order characteristics of Japanese. Chapter 3, "Fundamental constructions", starts with a detailed argument for the existence of a verb phrase (VP) in Japanese, and by corollary for the importance of subject, since VP abbreviates the category V[SUBCAT{PP[SBJ]}]. The rest of the book crucially exploits the notion of VP. The remainder of Chapter 3 examines the productive processes of causativization, passivization, and benefactivization. These phenomena have been the focus of continuing research since Kuroda 1965, and have generally been treated by considering the causative, passive, and benefactive morphemes as verbs taking sentential complements in the deep structure, with obligatory equi or raising and predicate raising. In Gunji's analysis, these morphemes subcategorize for subject, (possibly dative) object, and a (crucially) unsaturated verbal constituent. The latter is either a VP (= V[SUBCAT{PP[SBJ]}]), in causatives and intransitive (adversity) passives, or a TVP (= V[SUBCAT{PP[SBJ],PP[OBJ]}]) for the transitive passive. Given the unification-based semantics of the final chapter, the object control is captured simply by identifying the semantics of the object and the VP or TVP's subject in the lexical entries for the bound morphs. Gunji explicitly declines to give a consistent account of case marking in simplex sentences and these constructions, thus failing to explain why the causee may not be marked with accusative case when the embedded verb is transitive (the latter fact being captured with a feature co-occurrence restriction.) Neither does he appear to be worried by the existence of a phrase boundary occurring internally to a word, as in: Ken ga Naomi ni [vp hon wo yom] ase ta NOM DAT book ACC read CAUSE PAST (Ken made Naomi read the book.)